The present invention relates generally to a signal encoder and classifier, and more specifically to such an encoder and classifier for signal data obtained from a multiplicity of filters.
The assignee of this invention has developed certain systems for speech interpretation on which several patents have issued. The current invention involves several improvements and advances in regard to those patents. In addition, an additional application is currently pending which incorporates the technology disclosed in the prior patents, U.S. Pat. No. 3,812,291 and U.S. Pat. No. 3,582,559. These patents, as well as the pending application, Ser. No. 146,155, filed May 2, 1980, a continuation of Ser. No. 953,901, filed Oct. 23, 1978, and entitled Signal Pattern Encoder and Classifier, are incorporated herein by reference.
An automatic speech interpreter, as disclosed in the pending application, is essentially an acoustic pattern recognition device. Acoustically isolated utterances, such as words or phrases, are normalized by an information-theoretic compression technique that removes the effect of talker cadence and, to some degree, the effect of speaker variability. The resulting 120-bit pattern is then correlated with reference patterns derived through a training process. The only requirement for accurate recognition is reasonable acoustic separation between the patterns. The system can be retrained on line for new vocabularies, speakers, or acoustic environments.
Although the pending application describes an invention which has significant advances over the prior art in the areas of signal classification accuracy and economy in buffer storage space requirements, that application continues to experience a data buffering problem which is common to the field of art.
Upon recognition by the prior art devices that a complete word has been received, these devices discontinue reception of additional signal samples until the currently stored word is compressed, further encoded and tested. This halting of signal sampling results in a significant loss in signal continuity by the apparatus.
While the discontinuous sampling could theoretically be solved by serially buffering input data and later loading the buffered data, all in parallel, into a holding buffer, the volume of data normally required for speech and other signal pattern recognition is so large that such parallel data transfers become extremely costly. Furthermore, such data transfers are not common in micro-programmed implementations of such devices, and thus a viable solution to this problem has not been suggested in the prior art.
In addition, since the prior art typically shifts incoming data into a shift register, and then shifts this same data out of that register at the end of a word, the operations required for shifting the entire register contents must be accomplished for each word, regardless of the length of the word.